Datasets:
audio audioduration (s) 2.2 7.4 | id stringclasses 7
values | description stringclasses 7
values | expected_text stringclasses 7
values | transcribed_text stringclasses 7
values | sample_rate_hz int64 24k 24k | duration_ms int64 2.2k 7.4k | speech_start_ms int64 0 480 | speech_end_ms int64 1.9k 7.18k | trailing_non_speech_ms int64 200 1.3k | sha256 stringclasses 7
values | words listlengths 5 20 | generation dict |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
short_utterance | A short, complete spoken utterance. | Hello, this is a short turn. | Hello, this is a short turn. | 24,000 | 2,200 | 0 | 2,000 | 200 | 552dea44f88072ecc7f68a826138a24287bab90ed3e1d9e1708340ae4621ecb6 | [
{
"word": "Hello",
"start_ms": 0,
"end_ms": 460
},
{
"word": "this",
"start_ms": 1000,
"end_ms": 1080
},
{
"word": "is",
"start_ms": 1080,
"end_ms": 1240
},
{
"word": "a",
"start_ms": 1240,
"end_ms": 1480
},
{
"word": "short",
"start_ms": 1480,... | {
"speech_model": "gpt-4o-mini-tts",
"transcription_model": "whisper-1",
"voice": "alloy",
"instructions": "Speak naturally in clear American English at a steady conversational pace. Do not add words, sounds, or commentary.",
"timestamp_method": "openai_whisper_1_word_timestamps",
"timestamp_precision": "mo... | |
medium_utterance | A longer, two-sentence spoken utterance. | I am testing how the realtime API detects a complete spoken turn. This sentence should take several seconds to finish. | I am testing how the realtime API detects a complete spoken turn. This sentence should take several seconds to finish. | 24,000 | 7,400 | 480 | 7,180 | 220 | 788bd00b5abb30c6b23080b035ebe7e9a26cb8c5896e7cf7a72d98c3fcf06671 | [
{
"word": "I",
"start_ms": 480,
"end_ms": 780
},
{
"word": "am",
"start_ms": 780,
"end_ms": 980
},
{
"word": "testing",
"start_ms": 980,
"end_ms": 1380
},
{
"word": "how",
"start_ms": 1380,
"end_ms": 1680
},
{
"word": "the",
"start_ms": 1680,
... | {
"speech_model": "gpt-4o-mini-tts",
"transcription_model": "whisper-1",
"voice": "alloy",
"instructions": "Speak naturally in clear American English at a steady conversational pace. Do not add words, sounds, or commentary.",
"timestamp_method": "openai_whisper_1_word_timestamps",
"timestamp_precision": "mo... | |
short_statement | A concise declarative statement suitable for composing a multi-turn scenario. | The first turn ends here. | The first turn ends here. | 24,000 | 3,200 | 0 | 1,900 | 1,300 | e63e490c5bde7afbf12a76a2344fd30c728d39f6726f254e5f5e64ac027ae0ff | [
{
"word": "The",
"start_ms": 0,
"end_ms": 480
},
{
"word": "first",
"start_ms": 480,
"end_ms": 800
},
{
"word": "turn",
"start_ms": 800,
"end_ms": 1160
},
{
"word": "ends",
"start_ms": 1160,
"end_ms": 1560
},
{
"word": "here",
"start_ms": 1560,... | {
"speech_model": "gpt-4o-mini-tts",
"transcription_model": "whisper-1",
"voice": "alloy",
"instructions": "Speak naturally in clear American English at a steady conversational pace. Do not add words, sounds, or commentary.",
"timestamp_method": "openai_whisper_1_word_timestamps",
"timestamp_precision": "mo... | |
follow_up_statement | A second declarative statement suitable for composing a multi-turn scenario. | The second turn begins after the pause. | The second turn begins after the pause. | 24,000 | 2,950 | 0 | 2,360 | 590 | a598390e7dc0a3ba6d6ed4577402086e2253a2360ae7c72b682a51e793c0d0cd | [
{
"word": "The",
"start_ms": 0,
"end_ms": 540
},
{
"word": "second",
"start_ms": 540,
"end_ms": 820
},
{
"word": "turn",
"start_ms": 820,
"end_ms": 1060
},
{
"word": "begins",
"start_ms": 1060,
"end_ms": 1520
},
{
"word": "after",
"start_ms": 1... | {
"speech_model": "gpt-4o-mini-tts",
"transcription_model": "whisper-1",
"voice": "alloy",
"instructions": "Speak naturally in clear American English at a steady conversational pace. Do not add words, sounds, or commentary.",
"timestamp_method": "openai_whisper_1_word_timestamps",
"timestamp_precision": "mo... | |
diarization_speaker_a_first | The first utterance from speaker A in composed diarization scenarios. | The amber lighthouse marks the northern harbor, and I will return after the evening tide. | The amber lighthouse marks the northern harbor, and I will return after the evening tide. | 24,000 | 5,950 | 0 | 5,440 | 510 | 91732e4ebdd4991605b0699906a0b5d7f6f554e1475231feded69840cfdc23cf | [
{
"word": "The",
"start_ms": 0,
"end_ms": 400
},
{
"word": "amber",
"start_ms": 400,
"end_ms": 780
},
{
"word": "lighthouse",
"start_ms": 780,
"end_ms": 1380
},
{
"word": "marks",
"start_ms": 1380,
"end_ms": 1940
},
{
"word": "the",
"start_ms":... | {
"speech_model": "gpt-4o-mini-tts",
"transcription_model": "whisper-1",
"voice": "marin",
"instructions": "Speak naturally in clear American English at a steady conversational pace. Do not add words, sounds, or commentary.",
"timestamp_method": "openai_whisper_1_word_timestamps",
"timestamp_precision": "mo... | |
diarization_speaker_b | The utterance from speaker B in composed diarization scenarios. | Blue mountain trains leave from platform seven, while the station clock sounds twice. | Blue mountain trains leave from platform seven, while the station clock sounds twice. | 24,000 | 5,250 | 0 | 4,920 | 330 | c3a6ac324b49d28b05efaea2b397feab42ff6e27758540e618b841c6f098346c | [
{
"word": "Blue",
"start_ms": 0,
"end_ms": 840
},
{
"word": "mountain",
"start_ms": 840,
"end_ms": 1240
},
{
"word": "trains",
"start_ms": 1240,
"end_ms": 1580
},
{
"word": "leave",
"start_ms": 1580,
"end_ms": 1900
},
{
"word": "from",
"start_m... | {
"speech_model": "gpt-4o-mini-tts",
"transcription_model": "whisper-1",
"voice": "cedar",
"instructions": "Speak naturally in clear American English at a steady conversational pace. Do not add words, sounds, or commentary.",
"timestamp_method": "openai_whisper_1_word_timestamps",
"timestamp_precision": "mo... | |
diarization_speaker_a_last | The final utterance from speaker A in composed diarization scenarios. | The northern harbor is calm again, and the amber lighthouse remains visible from shore. | The northern harbor is calm again, and the amber lighthouse remains visible from shore. | 24,000 | 4,850 | 0 | 4,420 | 430 | 8dae44ba80b9a68f8b01af2cc50104b279c51fec2f7bbb68f01f13e332b758e7 | [
{
"word": "The",
"start_ms": 0,
"end_ms": 280
},
{
"word": "northern",
"start_ms": 280,
"end_ms": 620
},
{
"word": "harbor",
"start_ms": 620,
"end_ms": 920
},
{
"word": "is",
"start_ms": 920,
"end_ms": 1300
},
{
"word": "calm",
"start_ms": 1300... | {
"speech_model": "gpt-4o-mini-tts",
"transcription_model": "whisper-1",
"voice": "marin",
"instructions": "Speak naturally in clear American English at a steady conversational pace. Do not add words, sounds, or commentary.",
"timestamp_method": "openai_whisper_1_word_timestamps",
"timestamp_precision": "mo... |
Realtime speech test recordings
Synthetic speech recordings for black-box Realtime API behavior tests in Speaches. Each WAV file is the unmodified output of OpenAI text-to-speech. Tests are responsible for adding silence, combining recordings, and choosing streaming chunk boundaries for their scenarios.
metadata.jsonl follows the Hugging Face AudioFolder layout. Each record contains the generation inputs, file
digest, expected text, transcription, and word/speech intervals from a separate whisper-1 transcription.
The timing intervals are model-derived reference annotations, not sample-exact ground truth; tests should apply an
explicit tolerance. Consumers can use speech_start_ms and speech_end_ms to trim a recording when needed.
The checked-in generation manifest and script are the source of truth. Regeneration is procedural rather than bit-for-bit reproducible because the hosted speech and transcription models can vary between calls and releases.
- Downloads last month
- 80